The Effect of Innovation Box Regimes on Income Shifting and Real Activity

Transcription

1 The Effect of Innovation Box Regimes on Income Shifting and Real Activity Shannon Chen University of Arizona Michelle Hanlon Massachusetts Institute of Technology Lisa De Simone Stanford University Rebecca Lester Stanford University Preliminary and Incomplete: Please do not quote without permission July 2017 Abstract We study whether innovation box tax incentives, which reduce tax rates on innovation-related income, are associated with tax-motivated income shifting and local investment in the countries that implement these regimes. Using a matched sample of European multinationals subsidiaries operating in Europe, we find evidence consistent with firms engaging in less tax-motivated income shifting out of the country following the implementation of innovation boxes that provide the greatest tax benefits. We also find that innovation box regimes are associated with higher levels of fixed asset investment. Our study contributes to the patent box literature by evaluating outcomes that the literature has not previously examined and by informing the ongoing policy debate regarding the economic effects of innovation box regimes. We appreciate comments from Ken Klassen (discussant) and participants at the 2016 Stanford Accounting Summer Camp.

2 1. Introduction Concerned with shrinking fiscal revenues due to lower tax collections on reported income, many countries have implemented changes to their corporate tax systems. In particular, 17 countries have implemented a tax policy known as an innovation box, which permits income derived from the use of certain intellectual property (IP) to be taxed at a reduced rate. 1 The goal of these regimes has been to (i) retain mobile income that is otherwise shifted out of the country to lower-tax jurisdictions by multinational entities (MNEs) and possibly attract mobile income from higher-tax jurisdictions; (ii) increase innovation; and (iii) increase the amount of real economic activity within a country. Prior literature examines the effect of innovation box regimes on innovation, using measures of patent counts or patent quality, and finds that the introduction of an innovation box is positively associated with these measures. 2 In this paper, we study other potential outcomes of innovation box regimes: tax-motivated income shifting and local country investment. Prior literature on income shifting shows that income reported in a jurisdiction increases following a tax rate decrease (Grubert and Mutti, 1991). Thus, the lower tax rate on innovationrelated income in innovation box countries, as well as the evidence consistent with an increase in patents following the enactment of innovation box regimes, lead us to predict that less income is shifted out of innovation box countries (i.e., more income is retained within the country) following the implementation of the innovation box legislation. 3 However, the association between an 1 The incentive is referred to as an innovation box or a patent box because several countries include a box on the corporate income tax return for companies to report that they have income that qualifies for the lower rate (Sullivan, 2015). The 17 jurisdictions are: Belgium, Cyprus, France, Hungary, Ireland, Israel, Italy, Korea, Liechtenstein, Luxembourg, Malta, Netherlands, Portugal, Spain, the Nidwalden canton in Switzerland, Turkey, and the U.K. In July 2015, U.S. Representatives Charles Boustany and Richard Neal proposed a U.S. innovation box in the Innovation Promotion Act of See additional details of these innovation boxes in Section See Alstadsaeter et al. (2015), Bohm et al. (2015), Bradley et al. (2015), Ernst et al. (2014), Ernst and Spengel (2011), Griffith et al. (2014), Hassbring and Edwall (2013), and Karkinsky and Riedel (2012). 3 Statistical tests show that innovation box countries in western Europe (i.e., the countries in our sample) have statutory rates that are eight percent higher than non-innovation box European countries (p-value < 0.01). 1

3 innovation box incentive and tax-motivated income shifting in a particular jurisdiction is unclear ex ante. For example, firms generally file for patent protection in all countries in which they exploit an intangible (defined as performing further research and development (R&D) in the jurisdiction, conducting manufacturing in the jurisdiction, or even simply selling into the country); thus, it is not necessarily the case that the increase in patents would be reflected in different income shifting behavior. In addition, some argue that innovation box incentives are insufficient in that they may not reduce the applicable tax rate to a level lower than non-innovation box countries rates. 4 Therefore, it is an empirical question if firms will report more tax-motivated income in an innovation box country after an innovation box incentive is available. We examine this question on a sample of European subsidiaries (from Bureau van Dijk s Amadeus database) operating in Western Europe, including the fourteen European innovation box countries, aggregated at the firm-country-year level (e.g., Daimler-Spain-2008). 5 These affiliatelevel data permit a difference-in-difference research design in which we compare income shifting in the post-innovation box period to the level of income shifting pre-innovation box and to a matched sample of non-innovation box European affiliates. To test whether firms respond to innovation box incentives by altering income shifting behavior, we use an income prediction model from Hines and Rice (1994). The Hines and Rice (1994) model allows for tests of income shifting at the country-year level and has also been employed in recent income shifting studies (Dowd, Landefeld, and Moore, 2017; De Simone, 2016; Huizinga and Laeven, 2008; Markle, 2015). Because taxable income for the countries in our sample is unobservable, we use separate 4 For example, Merrill (2016) discusses the 2015 U.S. proposal to implement an innovation box, which would have resulted in an effective U.S. corporate tax rate reduction from 35 percent to 27.2 percent. He states that this decrease would be unlikely to achieve the objectives of retaining IP in the US and encouraging repatriation of IP migrated abroad. 5 We recognize that the Amadeus database has some known limitations. We discuss these in Section 3. 2

4 entity financial statement income as a proxy for the amount of reported taxable income. The staggered implementation of the innovation box regimes allows us to more confidently attribute any change in income shifting behavior in the post-innovation-box period to the innovation box tax incentives. We predict that the negative relation documented in Hines and Rice (1994) between reported income and the statutory tax rate is attenuated in our sample of innovation box jurisdictions (which are relatively high-tax countries) after the innovation box regime (and lower tax rate for innovation income) is in place. 6 We find evidence consistent with our prediction - reported income is less responsive to statutory tax rates in countries that provide the greatest innovation box tax benefit. Specifically, we estimate that, in countries that enacted the largest innovation-box tax benefits, a one percent change in the statutory tax rate is associated with a reduction of outbound income shifting of 9.81 percent relative to the matched control affiliates, or 230,927 higher profit for the average firm-country-year. A reduction of this outbound income shifting is consistent with the tax policy goals of attracting and/or retaining income in the innovation box country. In contrast, we find that firms operating in countries that offered relatively low benefits under the innovation box regime increase their income shifting activity after implementation of the tax benefit. Thus, the effectiveness of the innovation box regime hinges on whether the tax rate reduction is substantial. Second, we also test a real effect whether and to what extent innovation box regimes have attracted local country investment. Prior to the Organization for Economic Cooperation and Development s (OECD s) Base Erosion and Profit Shifting (BEPS) project, there were few to no 6 Ideally, we would directly test the relation between the innovation box tax rate and reported income. However, the amount of innovation box benefit is a function of the firm-country-year specific proportion of qualifying and nonqualifying innovation-related income, and data on a firm s innovation-related income are not available. Furthermore, there is no preferential rate within the control set of observations. Therefore, we test our first hypothesis by studying whether the responsiveness of reported profits to the statutory tax rate changes in the post-period. 3

5 requirements to locate real activity such as R&D or manufacturing in the jurisdiction of the innovation box to obtain the lower tax rate benefit. 7 However, we test this real outcome because much discussion about the benefit of innovation boxes is tied to an increased presence within a country. 8 For example, Evers et al. (2015) states whether the policies succeed in attracting real activities is likely to depend largely on the extent to which firms will choose to co-locate real activities alongside income streams the success of IP boxes in attracting real investments may therefore be limited First, firms may already be achieving low rates e.g. by shifting income to lower tax jurisdictions such that IP boxes are not as attractive as they may appear. Second, firms may respond to an IP box by moving paper profits but not the underlying real activity. Overall, whether IP boxes will be effective in attracting real innovative activities is unclear and will require empirical evidence. We test whether firms increase investment in the innovation box country by estimating the amount of affiliate-level capital expenditure spending in each year. Using this measure, we find evidence consistent with increased fixed asset investment after the implementation of an innovation box, and that this effect does not vary with the magnitude of the innovation box benefit. We estimate that innovation box firm-country-years have a percent change in capital expenditures that is 5.66 percentage points larger on average than for the matched sample of non-innovation box control firm-country-years. Given our sample average fixed assets of 4.8M, this translates 7 The BEPS project includes nexus guidelines that require some substantive activity within a country in order to claim an innovation box incentive. Because these guidelines were phased in starting June 30, 2016, most firms were not required to report associated investment during the period examined in this study to claim an incentive under the innovation box regime. However, the innovation box rules for the Netherlands, Portugal, and Spain did include some limited nexus requirements prior to We perform analyses of these countries in additional tests. 8 For example, in 2013 the U.K. Intellectual Property Office stated, Patents have a strong link to R&D and high-tech manufacturing and are used by innovative companies in a wide range of sectors [The innovation box regime] will provide an incentive for these companies in the UK to develop new innovative patented products. This will encourage companies to locate the high-value jobs associated with the development, manufacture, and exploitation of patents in the UK and maintain the UK s position as a world leader in patented technologies. 4

6 into 267,158 greater capital expenditures for innovation box firm-country-years after the implementation of the tax benefit relative to matched control firm-country-years. This result suggests that firms respond to innovation boxes by increasing firm investment in the local country, a result that is also consistent with the policy incentives of these regimes. This paper contributes to the existing literature by examining the economic effects of innovation box tax policies. We show that innovation box policies are associated with greater investment in fixed assets and with the reporting of a higher amount of income in jurisdictions that provide the largest innovation box benefits. By testing how firms respond to innovation box incentives, we contribute to the ongoing policy debate regarding economic effects of innovation box tax policies. These results inform policy-makers considering adoption of or changes to existing innovation box regimes. Furthermore, we contribute to academic research by showing that, when certain tax policies are adopted, measuring the responsiveness using statutory tax rates may not fully capture potential outcomes. In this setting, in which the applicable tax rate is reduced for qualifying innovation income, we demonstrate that firm responses such as income shifting and investment are captured by testing the extent to which firms are less responsive to the statutory tax rate. The paper proceeds as follows. Section 2 provides institutional details about innovation boxes, discusses the prior literature, and outlines the hypotheses. Section 3 discusses the research design and data, and Section 4 presents the findings. Section 5 concludes. 2. Innovation Boxes 2.1 Institutional Details An innovation box tax policy is one that applies a lower tax rate to a separate schedule or box of income. Some regimes are limited to patents, while others provide tax benefits to a wider range 5

7 of innovation and IP (we refer to all of the regimes as innovation boxes for simplicity). Innovation box regimes are back-end incentives, meaning tax benefits are awarded in the late stages of the innovation process. These incentives provide a reduced income tax rate for certain income arising from the exploitation of IP, generally through a 50 to 80 percent deduction or exemption of qualified IP income. In contrast, there are also front-end incentives, such as the U.S. R&D tax credit, that provide tax benefits during the early part of the innovation process when R&D is being conducted and when the outcome is more uncertain (Merrill et al., 2012; PwC, 2016). 9 There are 17 countries around the world with an innovation box in place during our sample period 2006 to Additional countries, including the U.S., have or are considering an innovation box regime. Because our data include affiliate-level data in Europe only, we discuss and test the fourteen European innovation boxes. The Appendix includes a summary of these regimes. The structure and details of innovation boxes vary across countries. For example, the rates of tax applicable to qualifying income range between 5 percent and 21 percent. Relative to non-innovation box statutory tax rates in these countries (10 percent to 35 percent), the innovation box rate represents a minimum of a 48 percent rate reduction on innovation-related income in Spain up to a maximum 100 percent reduction in Malta. The IP-related income that qualifies for the benefit also varies across jurisdictions. For example, qualifying income can be narrowly defined to only relate to patents and subsequent patent activity (as in Belgium), or it can be a much more inclusive definition that includes the industrial fabrication process ( imbedded IP), know-how, trademarks, copyrights, and software copyrights (as in Portugal). Requirements that development be performed domestically for resulting 9 There have been a number of studies on the effectiveness of these incentives (see Berger, 1993; Hall and Van Reenan, 2000; Rao, 2014; and Hoopes, 2015 for discussion of this literature). 6

8 innovation-related income to qualify for innovation box incentives vary by country and over time. Generally the Netherlands, Portugal and Spain have greater nexus requirements than the other countries. Finally, innovation box regimes also vary in allowing income related to pre-existing or acquired IP to qualify for these incentives. 2.2 Prior Literature The academic literature examining innovation box tax policies has primarily focused on testing if innovation, measured with patent application data, increases after an innovation box benefit is implemented in a country. For example, Karkinsky and Riedel (2009) regress the number of patent applications (measured at the subsidiary-year level) on the country-level tax rate for a sample of firms. 10 They find a negative association, which is interpreted as evidence that innovation box regimes affect the location of firms IP assets. Other studies (Ernst and Spengel, 2011; Hassbring and Edwall, 2013; Ernst et al., 2014; Alstadsaeter et al., 2015; Bohm et al., 2015; Bradley et al., 2015; Schwab and Todtenhaupt, 2016; Dudar and Voget, 2016) find similar associations. 11 Related studies use simulation methods to study the effect of tax rates and tax 10 The tax rate is measured three different ways: i) the country s statutory tax rate, ii) the tax rate differential between the local country and the rate of other affiliates in the same corporate group, and iii) a blended rate reflective of both the statutory rate and the withholding tax rate on royalties. 11 While all address the similar research question of how tax rates affect patent filings, the research designs in these studies differ. Ernst and Spengel (2011) estimate a probit model on a sample from 1998 to 2007 in which an indicator for having a patent (as well as the number of patents) is regressed on the corporate tax rate. Hassbring and Edwall (2013) use an indicator for innovation box countries rather than the tax rate as the independent variable of interest in studying a sample of firms from 2000 to Ernst et al. (2014) test the relation between patent quality (measured with number of forward citations and other measures) and the tax rate for a sample of firms from 1995 to 2007; Alstadsaeter et al. (2015) include both the corporate tax rate and an indicator for an innovation box country as variables of interest in studying firms from 2000 to Bohm et al. (2015) study the association between the likelihood of relocating a patent from an inventor s home country to a foreign country and the corporate income tax rate, as well as variables intended to capture the patent s quality and value (i.e., the number of countries in which the firm seeks patent protection, as well as the number of forward citations). Bradley et al. (2015) regress the number of patents on tax rates, an innovation box indicator, and the interaction of the two for a sample from 1990 to 2012 and also include tests of different ownership thresholds. Schwab and Todtenhaupt (2016) study cross-border changes in R&D activity (measured as the number of patent filings) within multinational firms with operations in innovation box jurisdictions, finding positive spillovers following the implementation of innovation box regimes without nexus requirements. Dudar and Voget (2016) compare tax elasticities of patents and trademarks and find that the location of trademarks is more sensitive than that of patents to changes in taxes. The collective evidence suggests that patent quality/number of 7

9 reforms on the location of firm IP (Griffith et al., 2010 and 2014) and estimate future cost of capital measures (Evers et al., 2015) for purposes of evaluating the benefit of these regimes. Notably, all of these studies focus on one potential outcome of these regimes innovation and operationalize this construct using patent data. Three concurrent papers extend the study of innovation box regimes to examine their effect on cross-border payments and the location of reported income. Bornemann and Osswald (2017) study firm responses to the Belgium innovation box. Using data from Bureau van Dijk, PATSTAT, and the National Bank of Belgium, they show that patent-owning firms do not significantly increase patenting activity following the implementation of the innovation box in Belgium but do report significantly lower effective tax rates. These lower tax rates are observed within a sub-sample of firms that they identify as unable to reduce their effective tax rates through tax-motivated income shifting domestic-only firms or multinationals without an immediate parent or subsidiary in a lower-tax jurisdiction. Two studies use U.S. data to study cross-border payments between U.S. entities and foreign related entities located in countries with innovation box incentives. Dowd, Landefeld, and Moore (2017) show that implementation of an innovation box regime in a foreign country results in a 50 percent increase in intangible payments from the U.S. subsidiary to a related entity located in that country. However, they also observe a corresponding decline in non-intangible payments, suggesting firms may be reclassifying (rather than increasing) payments out of the U.S. to qualify for preferential foreign innovation box tax treatment. Ohrn (2016) finds that U.S. payments to foreign parties for the use of intellectual property increase only in response to innovation box regimes that permit income from existing intellectual property to qualify, suggesting that specific patents/number of patent applications are negatively related to a country s tax rate (inclusive of any innovation box benefit). 8

10 elements of these regimes are important when testing firm responses. In our paper, we examine innovation box regimes across several countries and test whether and to what extent tax-motivated income shifting and investment are affected by these incentives. 2.3 Hypotheses We first examine the premise that firms respond to the innovation box incentive by changing existing income shifting behavior. Formally, our first hypothesis is as follows: H1: Tax-motivated reported income increases in innovation box firm-country-years following the implementation of an innovation box, relative to a matched sample of non-innovation box control firm-country-years. Although lower tax rates are generally associated with more reported income in the country, there are reasons why the innovation box regime may not yield similar results. Specifically, the regimes are more narrowly focused on IP income than other statutory rate changes and may not provide sufficiently lower tax rates than other countries basic corporate tax rate to attract additional income. For example, Ireland s corporate tax rate is 12.5 percent, and permitted tax structures such as the Double Irish effectively lowered the rate further. 12 In addition, although prior literature shows an increase in patent filings in innovation box countries following the implementation of the regime (e.g., Alstadsaeter et al. 2015, Bohm et al. 2015, Bradley et al. 2015, Ernst et al. 2014, Ernst and Spengel 2011, Griffith et al. 2014, Hassbring and Edwall 2013, and Karkinsky and Riedel 2012), the increase in patents may not be accompanied by an increase in income because firms generally file for patent protection in all jurisdictions in which they exploit an intangible asset. 12 The Irish Resident/Non-Resident (so-called Double Irish ) structure uses two corporations incorporated in Ireland, only one of which a tax resident of Ireland (the Irish Resident). The other corporation (the Irish Non-Resident), whose central management and control lies outside Ireland, typically has tax residency in a tax haven. The structure was popular among multinationals because it allowed them to shift profits, usually related to intangibles, to the Irish Non- Resident. Ireland disallowed the structure in 2015, although structures already in place at the time of the rule change are grandfathered until

11 Our second hypothesis examines whether innovation box regimes are associated with increased real activity within the jurisdiction. 13 Because prior literature shows that investment is negatively related to tax rates (see Hassett and Hubbard (2002) and Hassett and Newmark (2008) for a review of this literature), we expect an increase in investment following a reduction in a country s tax rate via the innovation box incentive. However, as noted above, the preferential innovation box regimes may be insufficient in attracting income, much less real activity. Consistent with prior literature, we predict the following: H2: Investment increases in innovation box firm-country-years following the implementation of an innovation box, relative to a matched sample of non-innovation box firm-countryyears. 3. Research Design and Data 3.1 Tests of H1 The accounting and economics literatures have yielded four main approaches to identifying tax-motivated income shifting, including Klassen, Lang, and Wolfson (1993); Hines and Rice (1994); Collins, Kemsley, and Lang (1998); and Dyreng and Markle (2016). Each of these approaches regresses a measure of firm performance (ROA, ROS, or the natural log of income) on a varying set of control variables and either a firm-level or country-level foreign tax rate. The Klassen et al. (1993), Collins et al. (1998), and Dyreng and Markle (2016) models all use consolidated financial statement data to study foreign versus domestic performance. In contrast, the Hines and Rice (1994) model uses country-year financial statement data of the subsidiaries of MNEs to more specifically study income shifting across the U.S. and multiple foreign jurisdictions. They use a Cobb-Douglas approach to estimate the amount of economic income 13 We also predict an increase in employment and R&D spending. We intend to study employment in future tests. However, due to data limitations, we are unable to test our prediction for R&D spending. 10

12 expected to be reported in a jurisdiction. This approach models pre-tax income as a function of the country s tax rate, as well as proxies for assets, labor, and productivity using aggregate countrylevel data. 14 We select the Hines and Rice (1994) model for several reasons. First, the Hines and Rice (1994) model is the only model originally estimated at the country level and, therefore, requires little alteration in our setting. Estimation at the country level permits more precise measurement of the constructs of interest (income shifting and investment), as well as the relation between these constructs and innovation box incentives, which also vary at the country level. In contrast, using the Klassen et al. (1993), Collins et al. (1998), or Dyreng and Markle (2016) models in our setting would require either consolidation of our country-level data or adaptation of the model for estimation at the country-year level. 15 Second, the dependent variable in the model is the log of pre-tax income and does not require scaling by other firm-specific measures that may be affected by innovation box regimes. For example, Klassen et al. (1993) use return on equity (ROE) and return on assets (ROA) as the dependent variable, which require scaling income by equity or assets, respectively, and Collins et al. (1998) use return on sales (ROS). 16 To the extent that any of these scalars are also affected by the innovation box tax incentive (for example, if total assets increase due to greater investment 14 Hines and Rice (1994) show a negative relation between income and tax rates: an increase of one percentage point in a country s tax rate is associated with a three percent reduction in reported profits. We test if this negative relation changes once lower rates on qualifying income are available under an innovation box incentive. 15 Further, the Dyreng and Markle (2016) model simultaneously estimates foreign and domestic profits. Adapting their model to the affiliate level would therefore require simultaneous equations for each jurisdiction in a given multinational group-year. The multinational groups in our sample own affiliates in up to 25 different western European jurisdictions in a given year, making this approach intractable in our setting. 16 Shareholder equity values of controlled affiliates can take on more extreme values relative to consolidated shareholder equity, which Klassen et al. (1993) use to attribute shareholder equity values to domestic versus foreign operations. Klassen et al. (1993) indicate that alternatively scaling by the book value of assets does not change the tenor of their results. 11

13 spending post-innovation box), then we may observe an empirical result that could be associated with changes in the scalar variable rather than by changes in reported income. Third, the model explicitly controls for levels of investment within a jurisdiction, which we also predict will be affected by an innovation box regime (H2). Because we expect to observe increased real activity in the post-innovation box period, and also because prior literature shows that investment is associated with the level of reported income, the empirical model we select must control for real activity effects. The Hines and Rice (1994) model is the only income shifting model that includes such controls. Finally, the model is commonly and recently used in other recent academic studies of tax-motivated income shifting (Huizinga and Laeven, 2008; Markle, 2015; De Simone, 2016). Though this model seems most appropriate for testing our hypotheses, we acknowledge that selection of this model imposes certain empirical trade-offs. First, the model assumes that the physical location of assets and labor can be used to predict economic income. Measures of physical assets generally exclude intangibles, which is problematic because intangibles are often used to facilitate cross-jurisdictional income shifting. Furthermore, intangibles are especially important in the context of studying firm responses to innovation-related incentives. However, to the extent tax incentives are a determinant of the location of intangibles, the use of intangibles to shift income should be captured by the tax rate incentive variable in the model. That is, inclusion of an intangibles variable in Eq. (1) would likely subsume some of the effects of the tax incentives. Another limitation of income shifting models is that they take the location of operations or other determinants of profitability as exogenous, despite the fact that a significant literature on location decisions shows that firm investment of physical and human capital responds to tax incentives. 12

14 We estimate the following Hines and Rice (1994) model, augmented with additional variables to test our income shifting hypothesis: Log(EBIT) i,c,t = α + β 1 *Log(Capital) i,c,t + β 2 *Log(Labor) i,c,t + β 3 *Log(GDP) c,t + β 4 *STR c,t + β 5 *Post i,c,t + β 6 *IB c + β 7 *STR c,t *Post i,c,t + β 8 *STR c,t *IB c +β 9 *Post i,c,t *IB c + β 10 *STR c,t *Post i,c,t *IB c + IndustryFE + YearFE (1) We estimate the model at the firm-country-year level, and the subscripts i, c, and t denote firm, country, and year, respectively. The dependent variable is the log of pre-tax income (EBIT). Capital i,c,t is tangible fixed assets, and Labor i,c,t is compensation expense. Consistent with prior literature, both capital and labor are predicted to be positively associated with reported income. Log(GDP) c,t is the log of GDP and controls for economic productivity, and we also expect a positive association between reported income and economic productivity. STR c,t is the statutory tax rate and thus does not reflect the lower innovation box rates. Because prior literature documents that reported income generally declines in response to high tax rates, we expect β 4 <0. Post i,c,t is an indicator variable equal to one for treatment and matched control firm-country-year observations in which the innovation box benefits are available for qualifying treatment firm-country-years, and zero otherwise. IB c is an indicator variable equal to one for firmcountry-years in countries that implement an innovation box regime and zero otherwise. Thus, our difference-in-differences design compares observations in an innovation box country after the innovation box was put in place (e.g., observations in the U.K. in 2014) to a matched sample of observations in non-innovation box countries over the same time period. We have no prediction for the coefficients on Post i,c,t, IB c, or their interaction, as it is unclear ex ante how the innovation box regime or countries implementing these benefits affect reported levels of income not otherwise explained by the determinants of the model (i.e., capital, labor, productivity, tax incentives, and time trends). In all specifications, we include industry and year fixed effects and cluster standard 13

15 errors by MNE group. We include year fixed effects, as well as the main effect of Post i,c,t, because the latter term is not a cross-sectional constant; that is, it varies by observation (in the case of matched control firm-country-years), country, and year in the sample. 17 β 10 is the coefficient of interest in testing H1. H1 predicts an increase in tax-motivated income reported in innovation box jurisdictions after the innovation box benefits are available, relative to a matched sample of non-innovation box firm-country-years. We test if income shifting behavior changes by examining if the relation between STR and Log(EBIT) is significantly different in the post-innovation box period for firm-country-years in innovation box countries, which have relatively high tax rates. Because STR c,t reflects the statutory tax rate for non-innovation box income, and because innovation boxes reduce this rate, we predict that reported income will be less sensitive to the statutory tax rate following the enactment of an innovation box regime, relative to other country-years. Showing an attenuation of this negative relation for innovation box countries in the post-innovation box period (i.e., β 10 >0) would be evidence consistent with firms shifting less income out of these relatively high-tax jurisdictions. 3.2 Tests of H2 To test our second hypothesis related to local country investment following the enactment of an innovation box, we estimate the following: PctChgCapital i,c,t = + β 1 *Log(TotalAssets) i,c,t-1 + β 2 *ROA i,c,t-1 + β 3 *SalesGrowth i,c,t-2 to t + β 4 *Leverage i,c,t-1 +β 5 *Log(GDP) c,t + β 6 *STR c,t + β 7 *Post i,c,t + β 8 *IB c + β 9 *Post i,c,t *IB c + IndustryFE + YearFE (2) The dependent variable PctChgCapital i,c,t estimates the amount of annual capital expenditure spending, which is not otherwise available in the Bureau van Dijk data. The numerator is the year- 17 In additional analyses, we re-estimate our results excluding year fixed effects. We discuss these results in Section 4. 14

16 on-year change in Capital i,c from year t-1 to t. We further adjust the numerator to add back depreciation expense (DEPR) in year t to reflect that the book value of fixed assets declines over time due to this expense (as opposed to fixed asset additions or disposals). We then scale this amount by Capital i,c,t Our variable of interest is the interaction between Post i,c,t and IB c (β 9 ) and captures the extent to which the level of investment differs in innovation-box countries in the years following implementation of the innovation box. 19 Consistent with firms responding to innovation box incentives by increasing real activity, we predict β 9 >0. We follow the literature (Desai, Foley, and Hines, 2004; Badertscher, Shroff, and White, 2013; Shroff, Verdi, and Yu, 2013; and Lester 2016) and include controls for size (Log(TotalAssets) i,c,t-1 ), performance (ROA i,c,t-1 and SalesGrowth i,c,t-2 to t ), debt (Leverage i,c,t-1 ), economic productivity (Log(GDP) c,t ), and tax incentives (STR c,t ). 3.3 Data and Samples Our samples begin with the Western European subsidiaries of European MNEs from Bureau van Dijk s Amadeus database. We focus on Western Europe because the region includes both innovation box and non-innovation box countries, and the implementation dates for the innovation box countries in the sample are staggered throughout the period, permitting more precise identification of the innovation box effects. Although we control for economic differences across countries using GDP data from the World Bank s World Development Indicators database, the 18 Capital expenditures (or an approximation thereof based on the change in assets) is also used in Almeida and Campello (2007), McNichols and Stubben (2008), Badertscher et al. (2013), and Shroff et al. (2014) to capture investment spending. In particular, Shroff et al. (2014) also use Bureau van Dijk data and calculate a similar measure of asset growth as their proxy for investment spending. 19 For H1, we test the effect of the regulatory change on income shifting using a triple interaction term. The triple interaction term captures the effect of income shifting (which itself is detected based on the coefficient on STR) following the implementation of an innovation box (Post) in innovation box countries (IB). Because we can directly observe investment and therefore are able to use a measure of investment as the dependent variable in Eq. (2) above, we test the effect of the regulatory change using the interaction term Post*IB. 15

17 relative homogeneity of economies also informs our decision to limit the sample to Western Europe. We obtain annual statutory tax rates (STR c,t ) from KPMG and aggregate subsidiary operations at the country-year level such that our unit of observation is firm-country-year. We present statutory tax rates in Table 1, where we also highlight country-years subject to an innovation box incentive. 20 We obtain consolidated financial statement information, ownership data, and separate entity financials for European MNEs from the Amadeus database. Limited liability entities are required to publicly report separate entity financials throughout Western Europe during our sample period, making this database relatively more complete for Western Europe as compared to other geographies available from Bureau van Dijk. However, we acknowledge there is no such reporting requirement for unlimited liability entities, such as Irish non-resident companies frequently used in the Double Irish structure. We use data from 2006 to 2014 for a starting sample of 77,450 firm-country-year observations with financial statement data required to estimate Equation (1). After retaining observations with industry codes and GDP data from the World Bank, we require each observation to report EBIT t- 1, Capital t-1 and Labor t-1 for purposes of matching treatment firms (i.e., firms operating in a country that implements an innovation box) to control firms (i.e., firms operating in a country that never implements an innovation box) in the year prior to enactment of the innovation box tax regime. We use nearest neighbor propensity score matching (with replacement) and match on lagged log of pre-tax income (EBIT) and log of tangible fixed assets (TFAS) within the same industry and 20 Table 1 presents the statutory tax rates that we use to estimate Equations (1) and (2). Note that the following countries are not included in our sample: Cyprus and Switzerland (no observations); France, Hungary, and Ireland (no observations in the period prior to the implementation of the innovation box regime); and Italy (no observations following the implementation of the innovation box regime). 16

18 year. This matching procedure results in 1,301 treatment-control pairs. The control observations reflect 1,078 distinct control firms, distributed across sixteen western European countries. Table 2 details the sample selection requirements. The H1 sample includes 15,155 firmcountry-years after imposing all necessary data restrictions. To test H2, we further require depreciation expense in year t to construct the dependent variable, as well as lagged total assets, total liabilities, and sales to construct our control variables. Additionally, we require that the dependent variable PctChgCapital i,c,t be within the range [-1, 5]. 21 Our resulting sample for the H2 tests is 13,665 firm-country-year observations. Table 3 provides descriptive statistics; for both samples, we provide summary statistics for EBIT (EBIT i,c,t ), total assets (TotalAssets i,c,t ), operating revenues (Sales i,c,t ), and tangible fixed assets (Capital i,c,t ). The unit of observation is the firm-country-year, and financial statement information is reported in thousands of Euros. On average, the firm-country-years in our samples are profitable (EBIT = 2.4 million) and large (average Total Assets = 25.6 million). These firmcountry-years report an average of 4.8 million in total fixed assets and 4.2 million in compensation expense. We also present summary statistics on variables used to estimate both models. For example, the average capital expenditure spending, captured by PctChgCapital, is equal to 42.6 percent of beginning-of-year fixed assets. The average amounts of ROA and sales growth for firms in the H2 sample are 12.9 percent and 41.8 percent, respectively. The mean statutory tax rates (STR c,t ) are approximately 26 percent in both samples. Untabulated univariate tests of differences in country statutory (non-innovation box) tax rates show that the average tax rate of 30.0 percent in innovation box countries is significantly higher 21 This range effectively limits capital expenditure spending to 500% of beginning-of-year fixed assets. This step results in losing 986 observations that report extremely high capital expenditure spending (up to 1,186%) relative to total fixed assets. 17

19 than the average statutory tax rate of 22.0 percent in non-innovation box countries (p<0.01). This finding suggests that our treatment observations face relatively high tax rates such that, absent innovation box incentives, they likely shift income out of the country to lower-tax related-party affiliates. An important identifying assumption in the difference-in-differences estimation is that treatment and control firms exhibit similar trends in the outcome variables prior to the passage of the innovation box regimes. Table 4 presents statistics that suggest that the treatment and control samples exhibit these parallel trends in the pre-periods. In Panel A, we test whether these two subsamples reflect different income shifting behavior in years t-3 to t-1. Because income shifting is captured as the coefficient on STR (rather than as a discrete dependent variable), we estimate the Hines & Rice (1994) model over these three years to study the pre-period parallel trends. We augment the model to include the indicator IB, as well as the interaction of IB and STR. To the extent that treatment and control firms do not exhibit parallel trends that is, if they exhibit different income shifting behavior in these earlier years we would observe a significant coefficient on the interaction term. Instead, we observe an insignificant coefficient (t=-0.09), which provides evidence consistent with these sub-samples having similar income shifting activity in the years leading up to the innovation box regimes. Panel B provides descriptive statistics related to the parallel trends assumption for H2. We first show the average amount of capital expenditure spending using PctChgCapital for the three-year period from t-3 to t-1. We also show the average capital expenditure spending for each pre-period year in the remaining columns. We compare these amounts to the average amounts for the control firms and observe no significant differences across any of these periods. This evidence is consistent 18

20 with treatment and control firms exhibiting similar trends in fixed asset spending in the years prior to the implementation of an innovation box regime. 4. Results of Empirical Tests 4.1 Tests of H1 Table 5 presents the results of testing H1. We begin by confirming that we can replicate inferences from the Hines and Rice (1994) model in our sample and present these results in Columns (1). Consistent with Hines and Rice (1994), we find a negative and significant coefficient on STR i,c,t, suggesting that firms report less profit in a country as tax rates increase. Also consistent with Hines and Rice (1994), we find a positive association between reported income and the proxies for capital, labor, and productivity. All results reported in Table 5 are robust to excluding year fixed effects. Columns (2) through (6) present results of testing H1. Recall that H1 predicts that firms will report more tax-motivated income in innovation box jurisdictions following the enactment of the reduced innovation box tax rate and that our variable of interest for testing H1 is the interaction between STR i,c,t, Post i,c,t, and IB c. Given that treatment observations in our sample experience high statutory tax rates relative to their matched control observations, a positive coefficient on this interaction is consistent with a reduction in the negative relation between profit and the statutory tax rate following the enactment of an innovation box, relative to non-innovation box firmcountry-years. In Column (2), we find no statistically significant coefficient on the triple interaction term. We then parse the sample based on statutory differences across the innovation box regimes. Columns (3) and (4) report results partitioned based on the magnitude of the innovation box 19

21 benefit. We calculate this benefit by subtracting a country s innovation box rate from the statutory tax rate and scaling by the statutory tax rate. 22 Column (3) reports results for estimating Eq. (1) on the treatment observations in countries that provide the greatest innovation box benefits (Belgium, Luxembourg, Malta, and the Netherlands), as well as their matched control observations. Column (4) reports results for the treatment observations in Portugal, Spain, and the United Kingdom (which provide relatively lower innovation box benefits), and their matched control observations. 23 We expect that relatively large rate reductions result in higher tax-motivated income reported in innovation box countries following the enactment of the innovation box regime. Consistent with this prediction, in Column (3) we find a positive and significant coefficient on the interaction between STR i,c,t, Post i,c,t, and IB c, suggesting reduced sensitivity of reported profits to the statutory tax rate in the subset of countries providing the greatest innovation box benefit. In contrast, we find a negative and significant coefficient on the triple interaction term in Column (4), suggesting an increase in tax-motivated income shifting behavior for the countries that provide the least innovation box benefit. In terms of economic magnitudes, the estimated coefficients from Column (3) mean that a one percent change in the statutory tax rate is associated with 9.81 percent higher reported in the highest-benefit innovation box countries relative to the matched control affiliates. Given the average level of EBIT, this is equivalent to 230,927 higher profit for the average firm-countryyear. Results in Column (4) suggest 3.76 percent lower reported profits in low-benefit innovation box countries, or 88,609 less profit for the average sample firm. These results highlight that the 22 For example, for a U.K. affiliate in 2014, we calculate a rate change of 56 percent: the 23 percent statutory tax rate in 2014 (Table 1) less the 10 percent innovation box rate in 2014 (Appendix), divided by the 23 percent statutory tax rate in 2014 (Table 1). 23 We note that the magnitude of the tax rate change is negatively correlated with how broadly each regime defines qualifying IP in our sample, suggesting that these two characteristics of innovation box benefits are used by policy makers as complements. 20

22 effectiveness of the innovation box regime in reducing tax-motivated income shifting hinges on the level of tax benefit provided. Columns (5) and (6) focus on whether the innovation box regime requires real activity in the jurisdiction for income to qualify for the lower innovation box tax rate. In Column (5), we include treatment observations in countries with such a nexus requirement (Netherlands, Portugal, and Spain) and their matched control observations. In Column (6) we include treatment observations in countries without a nexus requirement (Belgium, Luxembourg, Malta, and the United Kingdom) and their matched controls. We find a negative and significant coefficient on the interaction of interest in countries requiring nexus, indicating an increase in income shifting activity relative to the control sample. In Column (6), we find no difference in income shifting activity between treatment and control firms for countries without a nexus requirement. In terms of economic magnitudes, the average innovation box firm-country-year in jurisdictions requiring nexus reports 6.66 percent less EBIT in response to a one percent tax rate increase, or 156,792 less profits. One ex-post interpretation of this result is that the real activity requirement imposes too great a cost to a firm; consequently, the firm may forego the innovation box benefit and instead continue to shift income out of the country. In this case, the policy requiring nexus does not appear to achieve its intended response, likely due to the fact that other countries offer lower statutory rates and/or greater innovation box benefits without imposing a similar nexus requirement. 4.2 Tests of H2 Our multivariate tests of H2 model investment as a function of the Post i,c,t and IB c indicator variables, the interaction of these two indicator variables, and other determinants of investment. We report these results in Table 6. In Column (1), we first estimate the investment models without 21

23 Post i,c,t, IB c, or their interaction and find that fixed asset investment is increasing in firm profitability, sales growth, and leverage, but decreasing in size. We test H2 in columns (2) through (6). As discussed previously, the dependent variable PctChgCapital i,c,t is intended to approximate the annual spending on investment. In Column (2), we observe a positive and significant coefficient on the interaction between Post i,c,t and IB c, suggesting that investment increases on average in innovation box countries following the enactment of the innovation box. Specifically, the coefficient of on the interaction term suggests that the growth in fixed capital investment for innovation box firm-country-years from the pre- to post-innovation box period is 5.66 percentage points larger than for the matched sample of non-innovation box control firm-country-years. Given sample average fixed assets of 4.8M, this translates into 267,158 greater capital expenditures for innovation box firm-country-years after the implementation of the tax benefit relative to matched controls. The prior innovation box literature shows that the number of patents increases following the enactment of the innovation box regime; these results suggest that these intangibles are also accompanied by greater investments in fixed assets or plants in which firms products would be manufactured. Columns (3) and (4) examine cross-section differences in the magnitude of the innovation box tax benefit. As before, we calculate the innovation box rate reduction as the statutory tax rate less the maximum innovation box rate, scaled by the statutory tax rate. Column (3) reports results for high rate change countries Belgium, Luxembourg, Malta, and the Netherlands, and their matched control observations. Column (4) reports results for low rate change countries Portugal, Spain, and the United Kingdom, and their matched control observations. We find estimate positive, albeit insignificant coefficients, on the interaction term across both high and low rate changes. 22